AI Companies Can't Rely on 'Honor Code' for Oversight: Anthropic Policy Chief

Sarah Heck, Anthropic's head of public policy, expressed concerns about AI companies relying on an 'honor code' to manage oversight and safety. She stated that this approach is insufficient, saying 'We can't be checking our own homework.'

Heck made these comments at the Politico Decoded Summit in Washington, D.C., where she emphasized the need for collaboration with government to address AI's rapid development. This comes after Anthropic CEO Dario Amodei proposed a three-step plan to slow model capabilities without sacrificing commercial advantage.

Industry leaders like Sam Altman, Elon Musk, and Demis Hassabis have voiced support for Amodei's proposal. However, Nvidia's Jensen Huang and Meta's Mark Zuckerberg expressed skepticism about the need to intentionally slow development. Zuckerberg noted that Meta delayed its personal AI agent, Muse, by several months to focus on safety.

Heck stated that Anthropic engages in daily discussions with the White House and has been working with Congress throughout the year. She emphasized the importance of addressing hard questions now, such as AI regulation and safeguarding advancements.